openwrt/staging/blogic.git
5 years agoMerge tag 'kvmarm-fixes-for-5.1-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Tue, 30 Apr 2019 19:23:06 +0000 (21:23 +0200)]
Merge tag 'kvmarm-fixes-for-5.1-2' of git://git./linux/kernel/git/kvmarm/kvmarm into kvm-master

KVM/ARM fixes for 5.1, take #2:

- Don't try to emulate timers on userspace access
- Fix unaligned huge mappings, again
- Properly reset a vcpu that fails to reset(!)
- Properly retire pending LPIs on reset
- Fix computation of emulated CNTP_TVAL

5 years agoKVM: selftests: make hyperv_cpuid test pass on AMD
Vitaly Kuznetsov [Fri, 26 Apr 2019 13:27:11 +0000 (15:27 +0200)]
KVM: selftests: make hyperv_cpuid test pass on AMD

Enlightened VMCS is only supported on Intel CPUs but the test shouldn't
fail completely.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: lapic: Check for in-kernel LAPIC before deferencing apic pointer
Sean Christopherson [Fri, 26 Apr 2019 02:01:09 +0000 (19:01 -0700)]
KVM: lapic: Check for in-kernel LAPIC before deferencing apic pointer

...to avoid dereferencing a null pointer when querying the per-vCPU
timer advance.

Fixes: 39497d7660d98 ("KVM: lapic: Track lapic timer advance per vCPU")
Reported-by: syzbot+f7e65445a40d3e0e4ebf@syzkaller.appspotmail.com
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: fix KVM_CLEAR_DIRTY_LOG for memory slots of unaligned size
Paolo Bonzini [Wed, 17 Apr 2019 13:28:44 +0000 (15:28 +0200)]
KVM: fix KVM_CLEAR_DIRTY_LOG for memory slots of unaligned size

If a memory slot's size is not a multiple of 64 pages (256K), then
the KVM_CLEAR_DIRTY_LOG API is unusable: clearing the final 64 pages
either requires the requested page range to go beyond memslot->npages,
or requires log->num_pages to be unaligned, and kvm_clear_dirty_log_protect
requires log->num_pages to be both in range and aligned.

To allow this case, allow log->num_pages not to be a multiple of 64 if
it ends exactly on the last page of the slot.

Reported-by: Peter Xu <peterx@redhat.com>
Fixes: 98938aa8edd6 ("KVM: validate userspace input in kvm_clear_dirty_log_protect()", 2019-01-02)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agox86/kvm/mmu: reset MMU context when 32-bit guest switches PAE
Vitaly Kuznetsov [Tue, 30 Apr 2019 17:33:26 +0000 (19:33 +0200)]
x86/kvm/mmu: reset MMU context when 32-bit guest switches PAE

Commit 47c42e6b4192 ("KVM: x86: fix handling of role.cr4_pae and rename it
to 'gpte_size'") introduced a regression: 32-bit PAE guests stopped
working. The issue appears to be: when guest switches (enables) PAE we need
to re-initialize MMU context (set context->root_level, do
reset_rsvds_bits_mask(), ...) but init_kvm_tdp_mmu() doesn't do that
because we threw away is_pae(vcpu) flag from mmu role. Restore it to
kvm_mmu_extended_role (as we now don't need it in base role) to fix
the issue.

Fixes: 47c42e6b4192 ("KVM: x86: fix handling of role.cr4_pae and rename it to 'gpte_size'")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Whitelist port 0x7e for pre-incrementing %rip
Sean Christopherson [Mon, 29 Apr 2019 14:04:15 +0000 (07:04 -0700)]
KVM: x86: Whitelist port 0x7e for pre-incrementing %rip

KVM's recent bug fix to update %rip after emulating I/O broke userspace
that relied on the previous behavior of incrementing %rip prior to
exiting to userspace.  When running a Windows XP guest on AMD hardware,
Qemu may patch "OUT 0x7E" instructions in reaction to the OUT itself.
Because KVM's old behavior was to increment %rip before exiting to
userspace to handle the I/O, Qemu manually adjusted %rip to account for
the OUT instruction.

Arguably this is a userspace bug as KVM requires userspace to re-enter
the kernel to complete instruction emulation before taking any other
actions.  That being said, this is a bit of a grey area and breaking
userspace that has worked for many years is bad.

Pre-increment %rip on OUT to port 0x7e before exiting to userspace to
hack around the issue.

Fixes: 45def77ebf79e ("KVM: x86: update %rip after emulating IO")
Reported-by: Simon Becherer <simon@becherer.de>
Reported-and-tested-by: Iakov Karpov <srid@rkmail.ru>
Reported-by: Gabriele Balducci <balducci@units.it>
Reported-by: Antti Antinoja <reader@fennosys.fi>
Cc: stable@vger.kernel.org
Cc: Takashi Iwai <tiwai@suse.com>
Cc: Jiri Slaby <jslaby@suse.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoDocumentation: kvm: fix dirty log ioctl arch lists
Andrew Jones [Mon, 29 Apr 2019 08:27:10 +0000 (10:27 +0200)]
Documentation: kvm: fix dirty log ioctl arch lists

KVM_GET_DIRTY_LOG is implemented by all architectures, not just x86,
and KVM_CAP_MANUAL_DIRTY_LOG_PROTECT is additionally implemented by
arm, arm64, and mips.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Move RSB stuffing to before the first RET after VM-Exit
Rick Edgecombe [Sat, 27 Apr 2019 00:23:58 +0000 (17:23 -0700)]
KVM: VMX: Move RSB stuffing to before the first RET after VM-Exit

The not-so-recent change to move VMX's VM-Exit handing to a dedicated
"function" unintentionally exposed KVM to a speculative attack from the
guest by executing a RET prior to stuffing the RSB.  Make RSB stuffing
happen immediately after VM-Exit, before any unpaired returns.

Alternatively, the VM-Exit path could postpone full RSB stuffing until
its current location by stuffing the RSB only as needed, or by avoiding
returns in the VM-Exit path entirely, but both alternatives are beyond
ugly since vmx_vmexit() has multiple indirect callers (by way of
vmx_vmenter()).  And putting the RSB stuffing immediately after VM-Exit
makes it much less likely to be re-broken in the future.

Note, the cost of PUSH/POP could be avoided in the normal flow by
pairing the PUSH RAX with the POP RAX in __vmx_vcpu_run() and adding an
a POP to nested_vmx_check_vmentry_hw(), but such a weird/subtle
dependency is likely to cause problems in the long run, and PUSH/POP
will take all of a few cycles, which is peanuts compared to the number
of cycles required to fill the RSB.

Fixes: 453eafbe65f7 ("KVM: VMX: Move VM-Enter + VM-Exit handling to non-inline sub-routines")
Reported-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Co-developed-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: arm/arm64: Don't emulate virtual timers on userspace ioctls
Christoffer Dall [Thu, 25 Apr 2019 12:57:40 +0000 (13:57 +0100)]
KVM: arm/arm64: Don't emulate virtual timers on userspace ioctls

When a VCPU never runs before a guest exists, but we set timer registers
up via ioctls, the associated hrtimer might never get cancelled.

Since we moved vcpu_load/put into the arch-specific implementations and
only have load/put for KVM_RUN, we won't ever have a scheduled hrtimer
for emulating a timer when modifying the timer state via an ioctl from
user space.  All we need to do is make sure that we pick up the right
state when we load the timer state next time userspace calls KVM_RUN
again.

We also do not need to worry about this interacting with the bg_timer,
because if we were in WFI from the guest, and somehow ended up in a
kvm_arm_timer_set_reg, it means that:

 1. the VCPU thread has received a signal,
 2. we have called vcpu_load when being scheduled in again,
 3. we have called vcpu_put when we returned to userspace for it to issue
    another ioctl

And therefore will not have a bg_timer programmed and the event is
treated as a spurious wakeup from WFI if userspace decides to run the
vcpu again even if there are not virtual interrupts.

This fixes stray virtual timer interrupts triggered by an expiring
hrtimer, which happens after a failed live migration, for instance.

Fixes: bee038a674875 ("KVM: arm/arm64: Rework the timer code to use a timer_map")
Signed-off-by: Christoffer Dall <christoffer.dall@arm.com>
Reported-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
5 years agokvm: arm: Skip stage2 huge mappings for unaligned ipa backed by THP
Suzuki K Poulose [Wed, 10 Apr 2019 15:14:57 +0000 (16:14 +0100)]
kvm: arm: Skip stage2 huge mappings for unaligned ipa backed by THP

With commit a80868f398554842b14, we no longer ensure that the
THP page is properly aligned in the guest IPA. Skip the stage2
huge mapping for unaligned IPA backed by transparent hugepages.

Fixes: a80868f398554842b14 ("KVM: arm/arm64: Enforce PTE mappings at stage2 when needed")
Reported-by: Eric Auger <eric.auger@redhat.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Chirstoffer Dall <christoffer.dall@arm.com>
Cc: Zenghui Yu <yuzenghui@huawei.com>
Cc: Zheng Xiang <zhengxiang9@huawei.com>
Cc: Andrew Murray <andrew.murray@arm.com>
Cc: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
5 years agoKVM: arm/arm64: Ensure vcpu target is unset on reset failure
Andrew Jones [Thu, 4 Apr 2019 17:42:30 +0000 (19:42 +0200)]
KVM: arm/arm64: Ensure vcpu target is unset on reset failure

A failed KVM_ARM_VCPU_INIT should not set the vcpu target,
as the vcpu target is used by kvm_vcpu_initialized() to
determine if other vcpu ioctls may proceed. We need to set
the target before calling kvm_reset_vcpu(), but if that call
fails, we should then unset it and clear the feature bitmap
while we're at it.

Signed-off-by: Andrew Jones <drjones@redhat.com>
[maz: Simplified patch, completed commit message]
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
5 years agoKVM: lapic: Convert guest TSC to host time domain if necessary
Sean Christopherson [Wed, 17 Apr 2019 17:15:34 +0000 (10:15 -0700)]
KVM: lapic: Convert guest TSC to host time domain if necessary

To minimize the latency of timer interrupts as observed by the guest,
KVM adjusts the values it programs into the host timers to account for
the host's overhead of programming and handling the timer event.  In
the event that the adjustments are too aggressive, i.e. the timer fires
earlier than the guest expects, KVM busy waits immediately prior to
entering the guest.

Currently, KVM manually converts the delay from nanoseconds to clock
cycles.  But, the conversion is done in the guest's time domain, while
the delay occurs in the host's time domain.  This is perfectly ok when
the guest and host are using the same TSC ratio, but if the guest is
using a different ratio then the delay may not be accurate and could
wait too little or too long.

When the guest is not using the host's ratio, convert the delay from
guest clock cycles to host nanoseconds and use ndelay() instead of
__delay() to provide more accurate timing.  Because converting to
nanoseconds is relatively expensive, e.g. requires division and more
multiplication ops, continue using __delay() directly when guest and
host TSCs are running at the same ratio.

Cc: Liran Alon <liran.alon@oracle.com>
Cc: Wanpeng Li <wanpengli@tencent.com>
Cc: stable@vger.kernel.org
Fixes: 3b8a5df6c4dc6 ("KVM: LAPIC: Tune lapic_timer_advance_ns automatically")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: lapic: Allow user to disable adaptive tuning of timer advancement
Sean Christopherson [Wed, 17 Apr 2019 17:15:33 +0000 (10:15 -0700)]
KVM: lapic: Allow user to disable adaptive tuning of timer advancement

The introduction of adaptive tuning of lapic timer advancement did not
allow for the scenario where userspace would want to disable adaptive
tuning but still employ timer advancement, e.g. for testing purposes or
to handle a use case where adaptive tuning is unable to settle on a
suitable time.  This is epecially pertinent now that KVM places a hard
threshold on the maximum advancment time.

Rework the timer semantics to accept signed values, with a value of '-1'
being interpreted as "use adaptive tuning with KVM's internal default",
and any other value being used as an explicit advancement time, e.g. a
time of '0' effectively disables advancement.

Note, this does not completely restore the original behavior of
lapic_timer_advance_ns.  Prior to tracking the advancement per vCPU,
which is necessary to support autotuning, userspace could adjust
lapic_timer_advance_ns for *running* vCPU.  With per-vCPU tracking, the
module params are snapshotted at vCPU creation, i.e. applying a new
advancement effectively requires restarting a VM.

Dynamically updating a running vCPU is possible, e.g. a helper could be
added to retrieve the desired delay, choosing between the global module
param and the per-VCPU value depending on whether or not auto-tuning is
(globally) enabled, but introduces a great deal of complexity.  The
wrapper itself is not complex, but understanding and documenting the
effects of dynamically toggling auto-tuning and/or adjusting the timer
advancement is nigh impossible since the behavior would be dependent on
KVM's implementation as well as compiler optimizations.  In other words,
providing stable behavior would require extremely careful consideration
now and in the future.

Given that the expected use of a manually-tuned timer advancement is to
"tune once, run many", use the vastly simpler approach of recognizing
changes to the module params only when creating a new vCPU.

Cc: Liran Alon <liran.alon@oracle.com>
Cc: Wanpeng Li <wanpengli@tencent.com>
Reviewed-by: Liran Alon <liran.alon@oracle.com>
Cc: stable@vger.kernel.org
Fixes: 3b8a5df6c4dc6 ("KVM: LAPIC: Tune lapic_timer_advance_ns automatically")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: lapic: Track lapic timer advance per vCPU
Sean Christopherson [Wed, 17 Apr 2019 17:15:32 +0000 (10:15 -0700)]
KVM: lapic: Track lapic timer advance per vCPU

Automatically adjusting the globally-shared timer advancement could
corrupt the timer, e.g. if multiple vCPUs are concurrently adjusting
the advancement value.  That could be partially fixed by using a local
variable for the arithmetic, but it would still be susceptible to a
race when setting timer_advance_adjust_done.

And because virtual_tsc_khz and tsc_scaling_ratio are per-vCPU, the
correct calibration for a given vCPU may not apply to all vCPUs.

Furthermore, lapic_timer_advance_ns is marked __read_mostly, which is
effectively violated when finding a stable advancement takes an extended
amount of timer.

Opportunistically change the definition of lapic_timer_advance_ns to
a u32 so that it matches the style of struct kvm_timer.  Explicitly
pass the param to kvm_create_lapic() so that it doesn't have to be
exposed to lapic.c, thus reducing the probability of unintentionally
using the global value instead of the per-vCPU value.

Cc: Liran Alon <liran.alon@oracle.com>
Cc: Wanpeng Li <wanpengli@tencent.com>
Reviewed-by: Liran Alon <liran.alon@oracle.com>
Cc: stable@vger.kernel.org
Fixes: 3b8a5df6c4dc6 ("KVM: LAPIC: Tune lapic_timer_advance_ns automatically")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: lapic: Disable timer advancement if adaptive tuning goes haywire
Sean Christopherson [Wed, 17 Apr 2019 17:15:31 +0000 (10:15 -0700)]
KVM: lapic: Disable timer advancement if adaptive tuning goes haywire

To minimize the latency of timer interrupts as observed by the guest,
KVM adjusts the values it programs into the host timers to account for
the host's overhead of programming and handling the timer event.  Now
that the timer advancement is automatically tuned during runtime, it's
effectively unbounded by default, e.g. if KVM is running as L1 the
advancement can measure in hundreds of milliseconds.

Disable timer advancement if adaptive tuning yields an advancement of
more than 5000ns, as large advancements can break reasonable assumptions
of the guest, e.g. that a timer configured to fire after 1ms won't
arrive on the next instruction.  Although KVM busy waits to mitigate the
case of a timer event arriving too early, complications can arise when
shifting the interrupt too far, e.g. kvm-unit-test's vmx.interrupt test
will fail when its "host" exits on interrupts as KVM may inject the INTR
before the guest executes STI+HLT.   Arguably the unit test is "broken"
in the sense that delaying a timer interrupt by 1ms doesn't technically
guarantee the interrupt will arrive after STI+HLT, but it's a reasonable
assumption that KVM should support.

Furthermore, an unbounded advancement also effectively unbounds the time
spent busy waiting, e.g. if the guest programs a timer with a very large
delay.

5000ns is a somewhat arbitrary threshold.  When running on bare metal,
which is the intended use case, timer advancement is expected to be in
the general vicinity of 1000ns.  5000ns is high enough that false
positives are unlikely, while not being so high as to negatively affect
the host's performance/stability.

Note, a future patch will enable userspace to disable KVM's adaptive
tuning, which will allow priveleged userspace will to specifying an
advancement value in excess of this arbitrary threshold in order to
satisfy an abnormal use case.

Cc: Liran Alon <liran.alon@oracle.com>
Cc: Wanpeng Li <wanpengli@tencent.com>
Cc: stable@vger.kernel.org
Fixes: 3b8a5df6c4dc6 ("KVM: LAPIC: Tune lapic_timer_advance_ns automatically")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agox86: kvm: hyper-v: deal with buggy TLB flush requests from WS2012
Vitaly Kuznetsov [Wed, 20 Mar 2019 17:43:20 +0000 (18:43 +0100)]
x86: kvm: hyper-v: deal with buggy TLB flush requests from WS2012

It was reported that with some special Multi Processor Group configuration,
e.g:
 bcdedit.exe /set groupsize 1
 bcdedit.exe /set maxgroup on
 bcdedit.exe /set groupaware on
for a 16-vCPU guest WS2012 shows BSOD on boot when PV TLB flush mechanism
is in use.

Tracing kvm_hv_flush_tlb immediately reveals the issue:

 kvm_hv_flush_tlb: processor_mask 0x0 address_space 0x0 flags 0x2

The only flag set in this request is HV_FLUSH_ALL_VIRTUAL_ADDRESS_SPACES,
however, processor_mask is 0x0 and no HV_FLUSH_ALL_PROCESSORS is specified.
We don't flush anything and apparently it's not what Windows expects.

TLFS doesn't say anything about such requests and newer Windows versions
seem to be unaffected. This all feels like a WS2012 bug, which is, however,
easy to workaround in KVM: let's flush everything when we see an empty
flush request, over-flushing doesn't hurt.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Consider LAPIC TSC-Deadline timer expired if deadline too short
Liran Alon [Tue, 16 Apr 2019 17:36:34 +0000 (20:36 +0300)]
KVM: x86: Consider LAPIC TSC-Deadline timer expired if deadline too short

If guest sets MSR_IA32_TSCDEADLINE to value such that in host
time-domain it's shorter than lapic_timer_advance_ns, we can
reach a case that we call hrtimer_start() with expiration time set at
the past.

Because lapic_timer.timer is init with HRTIMER_MODE_ABS_PINNED, it
is not allowed to run in softirq and therefore will never expire.

To avoid such a scenario, verify that deadline expiration time is set on
host time-domain further than (now + lapic_timer_advance_ns).

A future patch can also consider adding a min_timer_deadline_ns module parameter,
similar to min_timer_period_us to avoid races that amount of ns it takes
to run logic could still call hrtimer_start() with expiration timer set
at the past.

Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoMerge tag 'kvm-ppc-fixes-5.1-1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Thu, 18 Apr 2019 16:53:12 +0000 (18:53 +0200)]
Merge tag 'kvm-ppc-fixes-5.1-1' of git://git./linux/kernel/git/paulus/powerpc into HEAD

KVM/PPC fixes for 5.1

- Fix host hang in the HTM assist code for POWER9
- Take srcu read lock around memslot lookup

5 years agoKVM: x86: avoid misreporting level-triggered irqs as edge-triggered in tracing
Vitaly Kuznetsov [Wed, 27 Mar 2019 14:12:20 +0000 (15:12 +0100)]
KVM: x86: avoid misreporting level-triggered irqs as edge-triggered in tracing

In __apic_accept_irq() interface trig_mode is int and actually on some code
paths it is set above u8:

kvm_apic_set_irq() extracts it from 'struct kvm_lapic_irq' where trig_mode
is u16. This is done on purpose as e.g. kvm_set_msi_irq() sets it to
(1 << 15) & e->msi.data

kvm_apic_local_deliver sets it to reg & (1 << 15).

Fix the immediate issue by making 'tm' into u16. We may also want to adjust
__apic_accept_irq() interface and use proper sizes for vector, level,
trig_mode but this is not urgent.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: fix spectrev1 gadgets
Paolo Bonzini [Thu, 11 Apr 2019 09:16:47 +0000 (11:16 +0200)]
KVM: fix spectrev1 gadgets

These were found with smatch, and then generalized when applicable.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: fix warning Using plain integer as NULL pointer
Hariprasad Kelam [Sat, 6 Apr 2019 09:36:58 +0000 (15:06 +0530)]
KVM: x86: fix warning Using plain integer as NULL pointer

Changed passing argument as "0 to NULL" which resolves below sparse warning

arch/x86/kvm/x86.c:3096:61: warning: Using plain integer as NULL pointer

Signed-off-by: Hariprasad Kelam <hariprasad.kelam@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm: add a selftest for SMM
Vitaly Kuznetsov [Wed, 10 Apr 2019 09:38:33 +0000 (11:38 +0200)]
selftests: kvm: add a selftest for SMM

Add a simple test for SMM, based on VMX.  The test implements its own
sync between the guest and the host as using our ucall library seems to
be too cumbersome: SMI handler is happening in real-address mode.

This patch also fixes KVM_SET_NESTED_STATE to happen after
KVM_SET_VCPU_EVENTS, in fact it places it last.  This is because
KVM needs to know whether the processor is in SMM or not.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm: fix for compilers that do not support -no-pie
Paolo Bonzini [Thu, 11 Apr 2019 13:51:19 +0000 (15:51 +0200)]
selftests: kvm: fix for compilers that do not support -no-pie

-no-pie was added to GCC at the same time as their configuration option
--enable-default-pie.  Compilers that were built before do not have
-no-pie, but they also do not need it.  Detect the option at build
time.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm/evmcs_test: complete I/O before migrating guest state
Paolo Bonzini [Thu, 11 Apr 2019 13:57:14 +0000 (15:57 +0200)]
selftests: kvm/evmcs_test: complete I/O before migrating guest state

Starting state migration after an IO exit without first completing IO
may result in test failures.  We already have two tests that need this
(this patch in fact fixes evmcs_test, similar to what was fixed for
state_test in commit 0f73bbc851ed, "KVM: selftests: complete IO before
migrating guest state", 2019-03-13) and a third is coming.  So, move the
code to vcpu_save_state, and while at it do not access register state
until after I/O is complete.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Always use 32-bit SMRAM save state for 32-bit kernels
Sean Christopherson [Tue, 2 Apr 2019 15:10:48 +0000 (08:10 -0700)]
KVM: x86: Always use 32-bit SMRAM save state for 32-bit kernels

Invoking the 64-bit variation on a 32-bit kenrel will crash the guest,
trigger a WARN, and/or lead to a buffer overrun in the host, e.g.
rsm_load_state_64() writes r8-r15 unconditionally, but enum kvm_reg and
thus x86_emulate_ctxt._regs only define r8-r15 for CONFIG_X86_64.

KVM allows userspace to report long mode support via CPUID, even though
the guest is all but guaranteed to crash if it actually tries to enable
long mode.  But, a pure 32-bit guest that is ignorant of long mode will
happily plod along.

SMM complicates things as 64-bit CPUs use a different SMRAM save state
area.  KVM handles this correctly for 64-bit kernels, e.g. uses the
legacy save state map if userspace has hid long mode from the guest,
but doesn't fare well when userspace reports long mode support on a
32-bit host kernel (32-bit KVM doesn't support 64-bit guests).

Since the alternative is to crash the guest, e.g. by not loading state
or explicitly requesting shutdown, unconditionally use the legacy SMRAM
save state map for 32-bit KVM.  If a guest has managed to get far enough
to handle SMIs when running under a weird/buggy userspace hypervisor,
then don't deliberately crash the guest since there are no downsides
(from KVM's perspective) to allow it to continue running.

Fixes: 660a5d517aaab ("KVM: x86: save/load state on SMM switch")
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Don't clear EFER during SMM transitions for 32-bit vCPU
Sean Christopherson [Tue, 2 Apr 2019 15:10:47 +0000 (08:10 -0700)]
KVM: x86: Don't clear EFER during SMM transitions for 32-bit vCPU

Neither AMD nor Intel CPUs have an EFER field in the legacy SMRAM save
state area, i.e. don't save/restore EFER across SMM transitions.  KVM
somewhat models this, e.g. doesn't clear EFER on entry to SMM if the
guest doesn't support long mode.  But during RSM, KVM unconditionally
clears EFER so that it can get back to pure 32-bit mode in order to
start loading CRs with their actual non-SMM values.

Clear EFER only when it will be written when loading the non-SMM state
so as to preserve bits that can theoretically be set on 32-bit vCPUs,
e.g. KVM always emulates EFER_SCE.

And because CR4.PAE is cleared only to play nice with EFER, wrap that
code in the long mode check as well.  Note, this may result in a
compiler warning about cr4 being consumed uninitialized.  Re-read CR4
even though it's technically unnecessary, as doing so allows for more
readable code and RSM emulation is not a performance critical path.

Fixes: 660a5d517aaab ("KVM: x86: save/load state on SMM switch")
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: clear SMM flags before loading state while leaving SMM
Sean Christopherson [Tue, 2 Apr 2019 15:03:11 +0000 (08:03 -0700)]
KVM: x86: clear SMM flags before loading state while leaving SMM

RSM emulation is currently broken on VMX when the interrupted guest has
CR4.VMXE=1.  Stop dancing around the issue of HF_SMM_MASK being set when
loading SMSTATE into architectural state, e.g. by toggling it for
problematic flows, and simply clear HF_SMM_MASK prior to loading
architectural state (from SMRAM save state area).

Reported-by: Jon Doron <arilou@gmail.com>
Cc: Jim Mattson <jmattson@google.com>
Cc: Liran Alon <liran.alon@oracle.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Fixes: 5bea5123cbf0 ("KVM: VMX: check nested state and CR4.VMXE against SMM")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Open code kvm_set_hflags
Sean Christopherson [Tue, 2 Apr 2019 15:03:10 +0000 (08:03 -0700)]
KVM: x86: Open code kvm_set_hflags

Prepare for clearing HF_SMM_MASK prior to loading state from the SMRAM
save state map, i.e. kvm_smm_changed() needs to be called after state
has been loaded and so cannot be done automatically when setting
hflags from RSM.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Load SMRAM in a single shot when leaving SMM
Sean Christopherson [Tue, 2 Apr 2019 15:03:09 +0000 (08:03 -0700)]
KVM: x86: Load SMRAM in a single shot when leaving SMM

RSM emulation is currently broken on VMX when the interrupted guest has
CR4.VMXE=1.  Rather than dance around the issue of HF_SMM_MASK being set
when loading SMSTATE into architectural state, ideally RSM emulation
itself would be reworked to clear HF_SMM_MASK prior to loading non-SMM
architectural state.

Ostensibly, the only motivation for having HF_SMM_MASK set throughout
the loading of state from the SMRAM save state area is so that the
memory accesses from GET_SMSTATE() are tagged with role.smm.  Load
all of the SMRAM save state area from guest memory at the beginning of
RSM emulation, and load state from the buffer instead of reading guest
memory one-by-one.

This paves the way for clearing HF_SMM_MASK prior to loading state,
and also aligns RSM with the enter_smm() behavior, which fills a
buffer and writes SMRAM save state in a single go.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: nVMX: Expose RDPMC-exiting only when guest supports PMU
Liran Alon [Mon, 25 Mar 2019 19:09:17 +0000 (21:09 +0200)]
KVM: nVMX: Expose RDPMC-exiting only when guest supports PMU

Issue was discovered when running kvm-unit-tests on KVM running as L1 on
top of Hyper-V.

When vmx_instruction_intercept unit-test attempts to run RDPMC to test
RDPMC-exiting, it is intercepted by L1 KVM which it's EXIT_REASON_RDPMC
handler raise #GP because vCPU exposed by Hyper-V doesn't support PMU.
Instead of unit-test expectation to be reflected with EXIT_REASON_RDPMC.

The reason vmx_instruction_intercept unit-test attempts to run RDPMC
even though Hyper-V doesn't support PMU is because L1 expose to L2
support for RDPMC-exiting. Which is reasonable to assume that is
supported only in case CPU supports PMU to being with.

Above issue can easily be simulated by modifying
vmx_instruction_intercept config in x86/unittests.cfg to run QEMU with
"-cpu host,+vmx,-pmu" and run unit-test.

To handle issue, change KVM to expose RDPMC-exiting only when guest
supports PMU.

Reported-by: Saar Amar <saaramar@microsoft.com>
Reviewed-by: Mihai Carabas <mihai.carabas@oracle.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Raise #GP when guest vCPU do not support PMU
Liran Alon [Mon, 25 Mar 2019 19:10:17 +0000 (21:10 +0200)]
KVM: x86: Raise #GP when guest vCPU do not support PMU

Before this change, reading a VMware pseduo PMC will succeed even when
PMU is not supported by guest. This can easily be seen by running
kvm-unit-test vmware_backdoors with "-cpu host,-pmu" option.

Reviewed-by: Mihai Carabas <mihai.carabas@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agox86/kvm: move kvm_load/put_guest_xcr0 into atomic context
WANG Chao [Fri, 12 Apr 2019 07:55:39 +0000 (15:55 +0800)]
x86/kvm: move kvm_load/put_guest_xcr0 into atomic context

guest xcr0 could leak into host when MCE happens in guest mode. Because
do_machine_check() could schedule out at a few places.

For example:

kvm_load_guest_xcr0
...
kvm_x86_ops->run(vcpu) {
  vmx_vcpu_run
    vmx_complete_atomic_exit
      kvm_machine_check
        do_machine_check
          do_memory_failure
            memory_failure
              lock_page

In this case, host_xcr0 is 0x2ff, guest vcpu xcr0 is 0xff. After schedule
out, host cpu has guest xcr0 loaded (0xff).

In __switch_to {
     switch_fpu_finish
       copy_kernel_to_fpregs
         XRSTORS

If any bit i in XSTATE_BV[i] == 1 and xcr0[i] == 0, XRSTORS will
generate #GP (In this case, bit 9). Then ex_handler_fprestore kicks in
and tries to reinitialize fpu by restoring init fpu state. Same story as
last #GP, except we get DOUBLE FAULT this time.

Cc: stable@vger.kernel.org
Signed-off-by: WANG Chao <chao.wang@ucloud.cn>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: svm: make sure NMI is injected after nmi_singlestep
Vitaly Kuznetsov [Wed, 3 Apr 2019 14:06:42 +0000 (16:06 +0200)]
KVM: x86: svm: make sure NMI is injected after nmi_singlestep

I noticed that apic test from kvm-unit-tests always hangs on my EPYC 7401P,
the hanging test nmi-after-sti is trying to deliver 30000 NMIs and tracing
shows that we're sometimes able to deliver a few but never all.

When we're trying to inject an NMI we may fail to do so immediately for
various reasons, however, we still need to inject it so enable_nmi_window()
arms nmi_singlestep mode. #DB occurs as expected, but we're not checking
for pending NMIs before entering the guest and unless there's a different
event to process, the NMI will never get delivered.

Make KVM_REQ_EVENT request on the vCPU from db_interception() to make sure
pending NMIs are checked and possibly injected.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agosvm/avic: Fix invalidate logical APIC id entry
Suthikulpanit, Suravee [Tue, 26 Mar 2019 03:57:37 +0000 (03:57 +0000)]
svm/avic: Fix invalidate logical APIC id entry

Only clear the valid bit when invalidate logical APIC id entry.
The current logic clear the valid bit, but also set the rest of
the bits (including reserved bits) to 1.

Fixes: 98d90582be2e ('svm: Fix AVIC DFR and LDR handling')
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoRevert "svm: Fix AVIC incomplete IPI emulation"
Suthikulpanit, Suravee [Wed, 20 Mar 2019 08:12:28 +0000 (08:12 +0000)]
Revert "svm: Fix AVIC incomplete IPI emulation"

This reverts commit bb218fbcfaaa3b115d4cd7a43c0ca164f3a96e57.

As Oren Twaig pointed out the old discussion:

  https://patchwork.kernel.org/patch/8292231/

that the change coud potentially cause an extra IPI to be sent to
the destination vcpu because the AVIC hardware already set the IRR bit
before the incomplete IPI #VMEXIT with id=1 (target vcpu is not running).
Since writting to ICR and ICR2 will also set the IRR. If something triggers
the destination vcpu to get scheduled before the emulation finishes, then
this could result in an additional IPI.

Also, the issue mentioned in the commit bb218fbcfaaa was misdiagnosed.

Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Reported-by: Oren Twaig <oren@scalemp.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: mmu: Fix overflow on kvm mmu page limit calculation
Ben Gardon [Mon, 8 Apr 2019 18:07:30 +0000 (11:07 -0700)]
kvm: mmu: Fix overflow on kvm mmu page limit calculation

KVM bases its memory usage limits on the total number of guest pages
across all memslots. However, those limits, and the calculations to
produce them, use 32 bit unsigned integers. This can result in overflow
if a VM has more guest pages that can be represented by a u32. As a
result of this overflow, KVM can use a low limit on the number of MMU
pages it will allocate. This makes KVM unable to map all of guest memory
at once, prompting spurious faults.

Tested: Ran all kvm-unit-tests on an Intel Haswell machine. This patch
introduced no new failures.

Signed-off-by: Ben Gardon <bgardon@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: nVMX: always use early vmcs check when EPT is disabled
Paolo Bonzini [Mon, 15 Apr 2019 13:57:19 +0000 (15:57 +0200)]
KVM: nVMX: always use early vmcs check when EPT is disabled

The remaining failures of vmx.flat when EPT is disabled are caused by
incorrectly reflecting VMfails to the L1 hypervisor.  What happens is
that nested_vmx_restore_host_state corrupts the guest CR3, reloading it
with the host's shadow CR3 instead, because it blindly loads GUEST_CR3
from the vmcs01.

For simplicity let's just always use hardware VMCS checks when EPT is
disabled.  This way, nested_vmx_restore_host_state is not reached at
all (or at least shouldn't be reached).

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: nVMX: allow tests to use bad virtual-APIC page address
Paolo Bonzini [Mon, 15 Apr 2019 13:16:17 +0000 (15:16 +0200)]
KVM: nVMX: allow tests to use bad virtual-APIC page address

As mentioned in the comment, there are some special cases where we can simply
clear the TPR shadow bit from the CPU-based execution controls in the vmcs02.
Handle them so that we can remove some XFAILs from vmx.flat.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86/mmu: Fix an inverted list_empty() check when zapping sptes
Sean Christopherson [Sat, 13 Apr 2019 02:55:41 +0000 (19:55 -0700)]
KVM: x86/mmu: Fix an inverted list_empty() check when zapping sptes

A recently introduced helper for handling zap vs. remote flush
incorrectly bails early, effectively leaking defunct shadow pages.
Manifests as a slab BUG when exiting KVM due to the shadow pages
being alive when their associated cache is destroyed.

==========================================================================
BUG kvm_mmu_page_header: Objects remaining in kvm_mmu_page_header on ...
--------------------------------------------------------------------------
Disabling lock debugging due to kernel taint
INFO: Slab 0x00000000fc436387 objects=26 used=23 fp=0x00000000d023caee ...
CPU: 6 PID: 4315 Comm: rmmod Tainted: G    B             5.1.0-rc2+ #19
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
Call Trace:
 dump_stack+0x46/0x5b
 slab_err+0xad/0xd0
 ? on_each_cpu_mask+0x3c/0x50
 ? ksm_migrate_page+0x60/0x60
 ? on_each_cpu_cond_mask+0x7c/0xa0
 ? __kmalloc+0x1ca/0x1e0
 __kmem_cache_shutdown+0x13a/0x310
 shutdown_cache+0xf/0x130
 kmem_cache_destroy+0x1d5/0x200
 kvm_mmu_module_exit+0xa/0x30 [kvm]
 kvm_arch_exit+0x45/0x60 [kvm]
 kvm_exit+0x6f/0x80 [kvm]
 vmx_exit+0x1a/0x50 [kvm_intel]
 __x64_sys_delete_module+0x153/0x1f0
 ? exit_to_usermode_loop+0x88/0xc0
 do_syscall_64+0x4f/0x100
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: a21136345cb6f ("KVM: x86/mmu: Split remote_flush+zap case out of kvm_mmu_flush_or_zap()")
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoBluetooth: btusb: request wake pin with NOAUTOEN
Brian Norris [Tue, 9 Apr 2019 18:49:17 +0000 (11:49 -0700)]
Bluetooth: btusb: request wake pin with NOAUTOEN

Badly-designed systems might have (for example) active-high wake pins
that default to high (e.g., because of external pull ups) until they
have an active firmware which starts driving it low.  This can cause an
interrupt storm in the time between request_irq() and disable_irq().

We don't support shared interrupts here, so let's just pre-configure the
interrupt to avoid auto-enabling it.

Fixes: fd913ef7ce61 ("Bluetooth: btusb: Add out-of-band wakeup support")
Fixes: 5364a0b4f4be ("arm64: dts: rockchip: move QCA6174A wakeup pin into its USB node")
Signed-off-by: Brian Norris <briannorris@chromium.org>
Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoMerge tag 'mips_fixes_5.1_2' of git://git.kernel.org/pub/scm/linux/kernel/git/mips...
Linus Torvalds [Wed, 10 Apr 2019 02:27:18 +0000 (16:27 -1000)]
Merge tag 'mips_fixes_5.1_2' of git://git./linux/kernel/git/mips/linux

Pull MIPS fixes from Paul Burton:
 "A few minor MIPS fixes:

   - Provide struct pt_regs * from get_irq_regs() to kgdb_nmicallback()
     when handling an IPI triggered by kgdb_roundup_cpus(), matching the
     behavior of other architectures & resolving kgdb issues for SMP
     systems.

   - Defer a pointer dereference until after a NULL check in the
     irq_shutdown callback for SGI IP27 HUB interrupts.

   - A defconfig update for the MSCC Ocelot to enable some necessary
     drivers"

* tag 'mips_fixes_5.1_2' of git://git.kernel.org/pub/scm/linux/kernel/git/mips/linux:
  MIPS: generic: Add switchdev, pinctrl and fit to ocelot_defconfig
  MIPS: SGI-IP27: Fix use of unchecked pointer in shutdown_bridge_irq
  MIPS: KGDB: fix kgdb support for SMP platforms.

5 years agoMerge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Wed, 10 Apr 2019 02:20:59 +0000 (16:20 -1000)]
Merge branch 'fixes' of git://git./linux/kernel/git/viro/vfs

Pull misc fixes from Al Viro:
 "A few regression fixes from this cycle"

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  aio: use kmem_cache_free() instead of kfree()
  iov_iter: Fix build error without CONFIG_CRYPTO
  aio: Fix an error code in __io_submit_one()

5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Tue, 9 Apr 2019 03:10:46 +0000 (17:10 -1000)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Off by one and bounds checking fixes in NFC, from Dan Carpenter.

 2) There have been many weird regressions in r8169 since we turned ASPM
    support on, some are still not understood nor completely resolved.
    Let's turn this back off for now. From Heiner Kallweit.

 3) Signess fixes for ethtool speed value handling, from Michael
    Zhivich.

 4) Handle timestamps properly in macb driver, from Paul Thomas.

 5) Two erspan fixes, it's the usual "skb ->data potentially reallocated
    and we're holding a stale protocol header pointer". From Lorenzo
    Bianconi.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  bnxt_en: Reset device on RX buffer errors.
  bnxt_en: Improve RX consumer index validity check.
  net: macb driver, check for SKBTX_HW_TSTAMP
  qlogic: qlcnic: fix use of SPEED_UNKNOWN ethtool constant
  broadcom: tg3: fix use of SPEED_UNKNOWN ethtool constant
  ethtool: avoid signed-unsigned comparison in ethtool_validate_speed()
  net: ip6_gre: fix possible use-after-free in ip6erspan_rcv
  net: ip_gre: fix possible use-after-free in erspan_rcv
  r8169: disable ASPM again
  MAINTAINERS: ieee802154: update documentation file pattern
  net: vrf: Fix ping failed when vrf mtu is set to 0
  selftests: add a tc matchall test case
  nfc: nci: Potential off by one in ->pipes[] array
  NFC: nci: Add some bounds checking in nci_hci_cmd_received()

5 years agoMerge branch 'fixes-v5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Tue, 9 Apr 2019 03:06:43 +0000 (17:06 -1000)]
Merge branch 'fixes-v5.1' of git://git./linux/kernel/git/jmorris/linux-security

Pull TPM fixes from James Morris:
 "From Jarkko: These are critical fixes for v5.1. Contains also couple
  of new selftests for v5.1 features (partial reads in /dev/tpm0)"

* 'fixes-v5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security:
  selftests/tpm2: Open tpm dev in unbuffered mode
  selftests/tpm2: Extend tests to cover partial reads
  KEYS: trusted: fix -Wvarags warning
  tpm: Fix the type of the return value in calc_tpm2_event_size()
  KEYS: trusted: allow trusted.ko to initialize w/o a TPM
  tpm: fix an invalid condition in tpm_common_poll
  tpm: turn on TPM on suspend for TPM 1.x

5 years agoMerge tag 'xtensa-20190408' of git://github.com/jcmvbkbc/linux-xtensa
Linus Torvalds [Tue, 9 Apr 2019 03:04:42 +0000 (17:04 -1000)]
Merge tag 'xtensa-20190408' of git://github.com/jcmvbkbc/linux-xtensa

Pull xtensa fixes from Max Filippov:

 - fix syscall number passed to trace_sys_exit

 - fix syscall number initialization in start_thread

 - fix level interpretation in the return_address

 - fix format string warning in init_pmd

* tag 'xtensa-20190408' of git://github.com/jcmvbkbc/linux-xtensa:
  xtensa: fix format string warning in init_pmd
  xtensa: fix return_address
  xtensa: fix initialization of pt_regs::syscall in start_thread
  xtensa: use actual syscall number in do_syscall_trace_leave

5 years agoMerge branch 'bnxt_en-fixes'
David S. Miller [Mon, 8 Apr 2019 23:39:41 +0000 (16:39 -0700)]
Merge branch 'bnxt_en-fixes'

Michael Chan says:

====================
bnxt_en: 2 bug fixes.

The first patch prevents possible driver crash if we get a bad RX index
from the hardware.  The second patch resets the device when the hardware
reports buffer error to recover from the error.

Please queue these for -stable also.  Thanks.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobnxt_en: Reset device on RX buffer errors.
Michael Chan [Mon, 8 Apr 2019 21:39:55 +0000 (17:39 -0400)]
bnxt_en: Reset device on RX buffer errors.

If the RX completion indicates RX buffers errors, the RX ring will be
disabled by firmware and no packets will be received on that ring from
that point on.  Recover by resetting the device.

Fixes: c0c050c58d84 ("bnxt_en: New Broadcom ethernet driver.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobnxt_en: Improve RX consumer index validity check.
Michael Chan [Mon, 8 Apr 2019 21:39:54 +0000 (17:39 -0400)]
bnxt_en: Improve RX consumer index validity check.

There is logic to check that the RX/TPA consumer index is the expected
index to work around a hardware problem.  However, the potentially bad
consumer index is first used to index into an array to reference an entry.
This can potentially crash if the bad consumer index is beyond legal
range.  Improve the logic to use the consumer index for dereferencing
after the validity check and log an error message.

Fixes: fa7e28127a5a ("bnxt_en: Add workaround to detect bad opaque in rx completion (part 2)")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: macb driver, check for SKBTX_HW_TSTAMP
Paul Thomas [Mon, 8 Apr 2019 19:37:54 +0000 (15:37 -0400)]
net: macb driver, check for SKBTX_HW_TSTAMP

Make sure SKBTX_HW_TSTAMP (i.e. SOF_TIMESTAMPING_TX_HARDWARE) has been
enabled for this skb. It does fix the issue where normal socks that
aren't expecting a timestamp will not wake up on select, but when a
user does want a SOF_TIMESTAMPING_TX_HARDWARE it does work.

Signed-off-by: Paul Thomas <pthomas8589@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ethtool-fix-use-of-SPEED_UNKNOWN-constant'
David S. Miller [Mon, 8 Apr 2019 23:30:43 +0000 (16:30 -0700)]
Merge branch 'ethtool-fix-use-of-SPEED_UNKNOWN-constant'

Michael Zhivich says:

====================
ethtool: fix use of SPEED_UNKNOWN constant

This patch series addresses 2 related issues:

1. ethtool_validate_speed() triggers a "signed-unsigned comparison"
warning due to type difference of SPEED_UNKNOWN constant (int)
and argument to ethtool_validate_speed (__u32).

2. some drivers use u16 storage for SPEED_UNKNOWN constant,
resulting in value truncation and thus failure to test against
SPEED_UNKNOWN correctly.

This revised series addresses several feedback comments:
- split up the patch in to series
- do not unnecessarily change drivers that use "int" storage
  for speed values
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoqlogic: qlcnic: fix use of SPEED_UNKNOWN ethtool constant
Michael Zhivich [Mon, 8 Apr 2019 14:48:47 +0000 (10:48 -0400)]
qlogic: qlcnic: fix use of SPEED_UNKNOWN ethtool constant

qlcnic driver uses u16 to store SPEED_UKNOWN ethtool constant,
which is defined as -1, resulting in value truncation and
thus incorrect test results against SPEED_UNKNOWN.

For example, the following test will print "False":

    u16 speed = SPEED_UNKNOWN;

    if (speed == SPEED_UNKNOWN)
        printf("True");
    else
        printf("False");

Change storage of speed to use u32 to avoid this issue.

Signed-off-by: Michael Zhivich <mzhivich@akamai.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobroadcom: tg3: fix use of SPEED_UNKNOWN ethtool constant
Michael Zhivich [Mon, 8 Apr 2019 14:48:46 +0000 (10:48 -0400)]
broadcom: tg3: fix use of SPEED_UNKNOWN ethtool constant

tg3 driver uses u16 to store SPEED_UKNOWN ethtool constant,
which is defined as -1, resulting in value truncation and
thus incorrect test results against SPEED_UNKNOWN.

For example, the following test will print "False":

u16 speed = SPEED_UNKNOWN;

if (speed == SPEED_UNKNOWN)
    printf("True");
else
    printf("False");

Change storage of speed to use u32 to avoid this issue.

Signed-off-by: Michael Zhivich <mzhivich@akamai.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoethtool: avoid signed-unsigned comparison in ethtool_validate_speed()
Michael Zhivich [Mon, 8 Apr 2019 14:48:45 +0000 (10:48 -0400)]
ethtool: avoid signed-unsigned comparison in ethtool_validate_speed()

When building C++ userspace code that includes ethtool.h
with "-Werror -Wall", g++ complains about signed-unsigned comparison in
ethtool_validate_speed() due to definition of SPEED_UNKNOWN as -1.

Explicitly cast SPEED_UNKNOWN to __u32 to match type of
ethtool_validate_speed() argument.

Signed-off-by: Michael Zhivich <mzhivich@akamai.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'erspan-use-after-free'
David S. Miller [Mon, 8 Apr 2019 23:16:47 +0000 (16:16 -0700)]
Merge branch 'erspan-use-after-free'

Lorenzo Bianconi says:

====================
fix possible use-after-free in erspan_v{4,6}

Similar to what I did in commit bb9bd814ebf0 ("ipv6: sit: reset ip
header pointer in ipip6_rcv"), fix possible use-after-free in
erspan_rcv and ip6erspan_rcv extracting tunnel metadata since the
packet can be 'uncloned' running __iptunnel_pull_header
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ip6_gre: fix possible use-after-free in ip6erspan_rcv
Lorenzo Bianconi [Sat, 6 Apr 2019 15:16:53 +0000 (17:16 +0200)]
net: ip6_gre: fix possible use-after-free in ip6erspan_rcv

erspan_v6 tunnels run __iptunnel_pull_header on received skbs to remove
erspan header. This can determine a possible use-after-free accessing
pkt_md pointer in ip6erspan_rcv since the packet will be 'uncloned'
running pskb_expand_head if it is a cloned gso skb (e.g if the packet has
been sent though a veth device). Fix it resetting pkt_md pointer after
__iptunnel_pull_header

Fixes: 1d7e2ed22f8d ("net: erspan: refactor existing erspan code")
Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ip_gre: fix possible use-after-free in erspan_rcv
Lorenzo Bianconi [Sat, 6 Apr 2019 15:16:52 +0000 (17:16 +0200)]
net: ip_gre: fix possible use-after-free in erspan_rcv

erspan tunnels run __iptunnel_pull_header on received skbs to remove
gre and erspan headers. This can determine a possible use-after-free
accessing pkt_md pointer in erspan_rcv since the packet will be 'uncloned'
running pskb_expand_head if it is a cloned gso skb (e.g if the packet has
been sent though a veth device). Fix it resetting pkt_md pointer after
__iptunnel_pull_header

Fixes: 1d7e2ed22f8d ("net: erspan: refactor existing erspan code")
Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests/tpm2: Open tpm dev in unbuffered mode
Tadeusz Struk [Tue, 12 Feb 2019 23:42:05 +0000 (15:42 -0800)]
selftests/tpm2: Open tpm dev in unbuffered mode

In order to have control over how many bytes are read or written
the device needs to be opened in unbuffered mode.

Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Tested-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: James Morris <james.morris@microsoft.com>
5 years agoselftests/tpm2: Extend tests to cover partial reads
Tadeusz Struk [Tue, 12 Feb 2019 23:42:10 +0000 (15:42 -0800)]
selftests/tpm2: Extend tests to cover partial reads

Three new tests added:
1. Send get random cmd, read header in 1st read, read the rest in second
   read - expect success
2. Send get random cmd, read only part of the response, send another
   get random command, read the response - expect success
3. Send get random cmd followed by another get random cmd, without
   reading the first response - expect the second cmd to fail with -EBUSY

Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Tested-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: James Morris <james.morris@microsoft.com>
5 years agoKEYS: trusted: fix -Wvarags warning
ndesaulniers@google.com [Mon, 22 Oct 2018 23:43:57 +0000 (16:43 -0700)]
KEYS: trusted: fix -Wvarags warning

Fixes the warning reported by Clang:
security/keys/trusted.c:146:17: warning: passing an object that
undergoes default
      argument promotion to 'va_start' has undefined behavior [-Wvarargs]
        va_start(argp, h3);
                       ^
security/keys/trusted.c:126:37: note: parameter of type 'unsigned
char' is declared here
unsigned char *h2, unsigned char h3, ...)
                               ^
Specifically, it seems that both the C90 (4.8.1.1) and C11 (7.16.1.4)
standards explicitly call this out as undefined behavior:

The parameter parmN is the identifier of the rightmost parameter in
the variable parameter list in the function definition (the one just
before the ...). If the parameter parmN is declared with ... or with a
type that is not compatible with the type that results after
application of the default argument promotions, the behavior is
undefined.

Link: https://github.com/ClangBuiltLinux/linux/issues/41
Link: https://www.eskimo.com/~scs/cclass/int/sx11c.html
Suggested-by: David Laight <David.Laight@aculab.com>
Suggested-by: Denis Kenzior <denkenz@gmail.com>
Suggested-by: James Bottomley <jejb@linux.vnet.ibm.com>
Suggested-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
Tested-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: James Morris <james.morris@microsoft.com>
5 years agotpm: Fix the type of the return value in calc_tpm2_event_size()
Yue Haibing [Wed, 20 Feb 2019 08:25:38 +0000 (16:25 +0800)]
tpm: Fix the type of the return value in calc_tpm2_event_size()

calc_tpm2_event_size() has an invalid signature because
it returns a 'size_t' where as its signature says that
it returns 'int'.

Cc: <stable@vger.kernel.org>
Fixes: 4d23cc323cdb ("tpm: add securityfs support for TPM 2.0 firmware event log")
Suggested-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: James Morris <james.morris@microsoft.com>
5 years agoKEYS: trusted: allow trusted.ko to initialize w/o a TPM
Jarkko Sakkinen [Mon, 25 Mar 2019 14:43:10 +0000 (16:43 +0200)]
KEYS: trusted: allow trusted.ko to initialize w/o a TPM

Allow trusted.ko to initialize w/o a TPM. This commit also adds checks
to the exported functions to fail when a TPM is not available.

Fixes: 240730437deb ("KEYS: trusted: explicitly use tpm_chip structure...")
Cc: James Morris <jmorris@namei.org>
Reported-by: Dan Williams <dan.j.williams@intel.com>
Tested-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: James Morris <james.morris@microsoft.com>
5 years agotpm: fix an invalid condition in tpm_common_poll
Tadeusz Struk [Wed, 27 Mar 2019 18:32:38 +0000 (11:32 -0700)]
tpm: fix an invalid condition in tpm_common_poll

The poll condition should only check response_length,
because reads should only be issued if there is data to read.
The response_read flag only prevents double writes.
The problem was that the write set the response_read to false,
enqued a tpm job, and returned. Then application called poll
which checked the response_read flag and returned EPOLLIN.
Then the application called read, but got nothing.
After all that the async_work kicked in.
Added also mutex_lock around the poll check to prevent
other possible race conditions.

Fixes: 9488585b21bef0df12 ("tpm: add support for partial reads")
Reported-by: Mantas Mikulėnas <grawity@gmail.com>
Tested-by: Mantas Mikulėnas <grawity@gmail.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: James Morris <james.morris@microsoft.com>
5 years agotpm: turn on TPM on suspend for TPM 1.x
Jarkko Sakkinen [Fri, 22 Mar 2019 10:51:20 +0000 (12:51 +0200)]
tpm: turn on TPM on suspend for TPM 1.x

tpm_chip_start/stop() should be also called for TPM 1.x devices on
suspend. Add that functionality back. Do not lock the chip because
it is unnecessary as there are no multiple threads using it when
doing the suspend.

Fixes: a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()")
Reported-by: Paul Zimmerman <pauldzim@gmail.com>
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Tested-by: Domenico Andreoli <domenico.andreoli@linux.com>
Signed-off-by: James Morris <james.morris@microsoft.com>
5 years agor8169: disable ASPM again
Heiner Kallweit [Fri, 5 Apr 2019 18:46:46 +0000 (20:46 +0200)]
r8169: disable ASPM again

There's a significant number of reports that re-enabling ASPM causes
different issues, ranging from decreased performance to system not
booting at all. This affects only a minority of users, but the number
of affected users is big enough that we better switch off ASPM again.

This will hurt notebook users who are not affected by the issues, they
may see decreased battery runtime w/o ASPM. With the PCI core folks is
being discussed to add generic sysfs attributes to control ASPM.
Once this is in place brave enough users can re-enable ASPM on their
system.

Fixes: a99790bf5c7f ("r8169: Reinstate ASPM Support")
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMAINTAINERS: ieee802154: update documentation file pattern
Stefan Schmidt [Mon, 8 Apr 2019 16:08:04 +0000 (18:08 +0200)]
MAINTAINERS: ieee802154: update documentation file pattern

When moving the documentation for the ieee802154 subsystem from
plain text to rst the file pattern in the MAINTAINERS file got wrong.
Updating it here to fix scripts using this file.

Reported-by: Joe Perches <joe@perches.com>
Signed-off-by: Stefan Schmidt <stefan@datenfreihafen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Mon, 8 Apr 2019 05:51:09 +0000 (19:51 -1000)]
Merge branch 'linus' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto fix from Herbert Xu:
 "This fixes a bug in the implementation of xcbc and cmac in caam"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: caam - fix copy of next buffer for xcbc and cmac

5 years agonet: vrf: Fix ping failed when vrf mtu is set to 0
Miaohe Lin [Mon, 8 Apr 2019 02:04:20 +0000 (10:04 +0800)]
net: vrf: Fix ping failed when vrf mtu is set to 0

When the mtu of a vrf device is set to 0, it would cause ping
failed. So I think we should limit vrf mtu in a reasonable range
to solve this problem. I set dev->min_mtu to IPV6_MIN_MTU, so it
will works for both ipv4 and ipv6. And if dev->max_mtu still be 0
can be confusing, so I set dev->max_mtu to ETH_MAX_MTU.

Here is the reproduce step:

1.Config vrf interface and set mtu to 0:
3: enp4s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel
master vrf1 state UP mode DEFAULT group default qlen 1000
    link/ether 52:54:00:9e:dd:c1 brd ff:ff:ff:ff:ff:ff

2.Ping peer:
3: enp4s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel
master vrf1 state UP group default qlen 1000
    link/ether 52:54:00:9e:dd:c1 brd ff:ff:ff:ff:ff:ff
    inet 10.0.0.1/16 scope global enp4s0
       valid_lft forever preferred_lft forever
connect: Network is unreachable

3.Set mtu to default value, ping works:
PING 10.0.0.2 (10.0.0.2) 56(84) bytes of data.
64 bytes from 10.0.0.2: icmp_seq=1 ttl=64 time=1.88 ms

Fixes: ad49bc6361ca2 ("net: vrf: remove MTU limits for vrf device")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoslab: fix a crash by reading /proc/slab_allocators
Qian Cai [Sat, 6 Apr 2019 22:59:01 +0000 (18:59 -0400)]
slab: fix a crash by reading /proc/slab_allocators

The commit 510ded33e075 ("slab: implement slab_root_caches list")
changes the name of the list node within "struct kmem_cache" from "list"
to "root_caches_node", but leaks_show() still use the "list" which
causes a crash when reading /proc/slab_allocators.

You need to have CONFIG_SLAB=y and CONFIG_MEMCG=y to see the problem,
because without MEMCG all slab caches are root caches, and the "list"
node happens to be the right one.

Fixes: 510ded33e075 ("slab: implement slab_root_caches list")
Signed-off-by: Qian Cai <cai@lca.pw>
Reviewed-by: Tobin C. Harding <tobin@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoselftests: add a tc matchall test case
Nicolas Dichtel [Fri, 5 Apr 2019 12:20:24 +0000 (14:20 +0200)]
selftests: add a tc matchall test case

This is a follow up of the commit 0db6f8befc32 ("net/sched: fix ->get
helper of the matchall cls").

To test it:
$ cd tools/testing/selftests/tc-testing
$ ln -s ../plugin-lib/nsPlugin.py plugins/20-nsPlugin.py
$ ./tdc.py -n -e 2638

Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoLinux 5.1-rc4
Linus Torvalds [Mon, 8 Apr 2019 00:09:59 +0000 (14:09 -1000)]
Linux 5.1-rc4

5 years agoMerge tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
Linus Torvalds [Sun, 7 Apr 2019 23:46:17 +0000 (13:46 -1000)]
Merge tag 'armsoc-fixes' of git://git./linux/kernel/git/soc/soc

Pull ARM SoC fixes from Olof Johansson:
 "A collection of fixes from the last few weeks. Most of them are
  smaller tweaks and fixes to DT and hardware descriptions for boards.
  Some of the more significant ones are:

   - eMMC and RGMII stability tweaks for rk3288

   - DDC fixes for Rock PI 4

   - Audio fixes for two TI am335x eval boards

   - D_CAN clock fix for am335x

   - Compilation fixes for clang

   - !HOTPLUG_CPU compilation fix for one of the new platforms this
     release (milbeaut)

   - A revert of a gpio fix for nomadik that instead was fixed in the
     gpio subsystem

   - Whitespace fix for the DT JSON schema (no tabs allowed)"

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc: (25 commits)
  ARM: milbeaut: fix build with !CONFIG_HOTPLUG_CPU
  ARM: iop: don't use using 64-bit DMA masks
  ARM: orion: don't use using 64-bit DMA masks
  Revert "ARM: dts: nomadik: Fix polarity of SPI CS"
  dt-bindings: cpu: Fix JSON schema
  arm/mach-at91/pm : fix possible object reference leak
  ARM: dts: at91: Fix typo in ISC_D0 on PC9
  ARM: dts: Fix dcan clkctrl clock for am3
  reset: meson-audio-arb: Fix missing .owner setting of reset_controller_dev
  dt-bindings: reset: meson-g12a: Add missing USB2 PHY resets
  ARM: dts: rockchip: Remove #address/#size-cells from rk3288-veyron gpio-keys
  ARM: dts: rockchip: Remove #address/#size-cells from rk3288 mipi_dsi
  ARM: dts: rockchip: Fix gpu opp node names for rk3288
  ARM: dts: am335x-evmsk: Correct the regulators for the audio codec
  ARM: dts: am335x-evm: Correct the regulators for the audio codec
  ARM: OMAP2+: add missing of_node_put after of_device_is_available
  ARM: OMAP1: ams-delta: Fix broken GPIO ID allocation
  arm64: dts: stratix10: add the sysmgr-syscon property from the gmac's
  arm64: dts: rockchip: fix rk3328 sdmmc0 write errors
  arm64: dts: rockchip: fix rk3328 rgmii high tx error rate
  ...

5 years agoMerge tag 'for-linus-20190407' of git://git.kernel.dk/linux-block
Linus Torvalds [Sun, 7 Apr 2019 23:28:36 +0000 (13:28 -1000)]
Merge tag 'for-linus-20190407' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:

 - Fixups for the pf/pcd queue handling (YueHaibing)

 - Revert of the three direct issue changes as they have been proven to
   cause an issue with dm-mpath (Bart)

 - Plug rq_count reset fix (Dongli)

 - io_uring double free in fileset registration error handling (me)

 - Make null_blk handle bad numa node passed in (John)

 - BFQ ifdef fix (Konstantin)

 - Flush queue leak fix (Shenghui)

 - Plug trace fix (Yufen)

* tag 'for-linus-20190407' of git://git.kernel.dk/linux-block:
  xsysace: Fix error handling in ace_setup
  null_blk: prevent crash from bad home_node value
  block: Revert v5.0 blk_mq_request_issue_directly() changes
  paride/pcd: Fix potential NULL pointer dereference and mem leak
  blk-mq: do not reset plug->rq_count before the list is sorted
  paride/pf: Fix potential NULL pointer dereference
  io_uring: fix double free in case of fileset regitration failure
  blk-mq: add trace block plug and unplug for multiple queues
  block: use blk_free_flush_queue() to free hctx->fq in blk_mq_init_hctx
  block/bfq: fix ifdef for CONFIG_BFQ_GROUP_IOSCHED=y

5 years agoARM: milbeaut: fix build with !CONFIG_HOTPLUG_CPU
Arnd Bergmann [Wed, 13 Mar 2019 21:19:16 +0000 (22:19 +0100)]
ARM: milbeaut: fix build with !CONFIG_HOTPLUG_CPU

When HOTPLUG_CPU is disabled, some fields in the smp operations
are not available or needed:

arch/arm/mach-milbeaut/platsmp.c:90:3: error: field designator 'cpu_die' does not refer to any field in type
      'struct smp_operations'
        .cpu_die                = m10v_cpu_die,
         ^
arch/arm/mach-milbeaut/platsmp.c:91:3: error: field designator 'cpu_kill' does not refer to any field in type
      'struct smp_operations'
        .cpu_kill               = m10v_cpu_kill,
         ^

Hide them in an #ifdef like the other platforms do.

Fixes: 9fb29c734f9e ("ARM: milbeaut: Add basic support for Milbeaut m10v SoC")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoARM: iop: don't use using 64-bit DMA masks
Arnd Bergmann [Mon, 25 Mar 2019 15:50:43 +0000 (16:50 +0100)]
ARM: iop: don't use using 64-bit DMA masks

clang warns about statically defined DMA masks from the DMA_BIT_MASK
macro with length 64:

 arch/arm/mach-iop13xx/setup.c:303:35: error: shift count >= width of type [-Werror,-Wshift-count-overflow]
 static u64 iop13xx_adma_dmamask = DMA_BIT_MASK(64);
                                  ^~~~~~~~~~~~~~~~
 include/linux/dma-mapping.h:141:54: note: expanded from macro 'DMA_BIT_MASK'
 #define DMA_BIT_MASK(n) (((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))
                                                      ^ ~~~

The ones in iop shouldn't really be 64 bit masks, so changing them
to what the driver can support avoids the warning.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoARM: orion: don't use using 64-bit DMA masks
Arnd Bergmann [Mon, 25 Mar 2019 15:50:42 +0000 (16:50 +0100)]
ARM: orion: don't use using 64-bit DMA masks

clang warns about statically defined DMA masks from the DMA_BIT_MASK
macro with length 64:

arch/arm/plat-orion/common.c:625:29: error: shift count >= width of type [-Werror,-Wshift-count-overflow]
                .coherent_dma_mask      = DMA_BIT_MASK(64),
                                          ^~~~~~~~~~~~~~~~
include/linux/dma-mapping.h:141:54: note: expanded from macro 'DMA_BIT_MASK'
 #define DMA_BIT_MASK(n) (((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))

The ones in orion shouldn't really be 64 bit masks, so changing them
to what the driver can support avoids the warning.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoRevert "ARM: dts: nomadik: Fix polarity of SPI CS"
Olof Johansson [Sun, 7 Apr 2019 22:18:41 +0000 (15:18 -0700)]
Revert "ARM: dts: nomadik: Fix polarity of SPI CS"

This reverts commit fa9463564e77067df81b0b8dec91adbbbc47bfb4.

Per Linus Walleij:

Dear ARM SoC maintainers,

can you please revert this patch. It was the wrong solution to the
wrong problem, and I must have acted in stress. Andrey fixed the
real bug in a proper way in these commits:

commit e5545c94e43b8f6599ffc01df8d1aedf18ee912a
"gpio: of: Check propname before applying "cs-gpios" quirks"
commit 7ce40277bf848391705011ba37eac2e377cbd9e6
"gpio: of: Check for "spi-cs-high" in child instead of parent node"

Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoMerge tag 'omap-for-v5.1/fixes-signed' of git://git.kernel.org/pub/scm/linux/kernel...
Olof Johansson [Sun, 7 Apr 2019 22:16:38 +0000 (15:16 -0700)]
Merge tag 'omap-for-v5.1/fixes-signed' of git://git./linux/kernel/git/tmlind/linux-omap into arm/fixes

Fixes for omaps for v5.1-rc cycle

Few small fixes for omap variants:

- Fix ams-delta gpio IDs
- Add missing of_node_put for omapdss platform init code
- Fix unconfigured audio regulators for two am335x boards
- Fix use of wrong offset for am335x d_can clocks

* tag 'omap-for-v5.1/fixes-signed' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap:
  ARM: dts: Fix dcan clkctrl clock for am3
  ARM: dts: am335x-evmsk: Correct the regulators for the audio codec
  ARM: dts: am335x-evm: Correct the regulators for the audio codec
  ARM: OMAP2+: add missing of_node_put after of_device_is_available
  ARM: OMAP1: ams-delta: Fix broken GPIO ID allocation

Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoMerge tag 'at91-5.1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/at91...
Olof Johansson [Sun, 7 Apr 2019 22:16:09 +0000 (15:16 -0700)]
Merge tag 'at91-5.1-fixes' of git://git./linux/kernel/git/at91/linux into arm/fixes

AT91 fixes for 5.1

- fix a typo in sama5d2 pinmuxing which concerns the ISC data 0 signal
- fix a kobject reference leak

* tag 'at91-5.1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/at91/linux:
  arm/mach-at91/pm : fix possible object reference leak
  ARM: dts: at91: Fix typo in ISC_D0 on PC9

Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoMerge tag 'v5.1-rockchip-dtfixes-1' of git://git.kernel.org/pub/scm/linux/kernel...
Olof Johansson [Sun, 7 Apr 2019 22:15:31 +0000 (15:15 -0700)]
Merge tag 'v5.1-rockchip-dtfixes-1' of git://git./linux/kernel/git/mmind/linux-rockchip into arm/fixes

Fixes for dtc warnings, fixes for ethernet transfers on rk3328,
sd-card related fixes on both rk3328 ans rk3288-tinker and a
regulator fix on rock64 and making ddc actually work on the
Rock PI 4 due to missing the ddc bus.

* tag 'v5.1-rockchip-dtfixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git/mmind/linux-rockchip:
  ARM: dts: rockchip: Remove #address/#size-cells from rk3288-veyron gpio-keys
  ARM: dts: rockchip: Remove #address/#size-cells from rk3288 mipi_dsi
  ARM: dts: rockchip: Fix gpu opp node names for rk3288
  arm64: dts: rockchip: fix rk3328 sdmmc0 write errors
  arm64: dts: rockchip: fix rk3328 rgmii high tx error rate
  ARM: dts: rockchip: Fix SD card detection on rk3288-tinker
  arm64: dts: rockchip: Fix vcc_host1_5v GPIO polarity on rk3328-rock64
  ARM: dts: rockchip: fix rk3288 cpu opp node reference
  arm64: dts: rockchip: add DDC bus on Rock Pi 4
  arm64: dts: rockchip: fix rk3328-roc-cc gmac2io tx/rx_delay

Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoMerge tag 'stratix10_fix_for_v5.1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Olof Johansson [Sun, 7 Apr 2019 22:14:30 +0000 (15:14 -0700)]
Merge tag 'stratix10_fix_for_v5.1' of git://git./linux/kernel/git/dinguyen/linux into arm/fixes

arm64: dts: stratix10: fix emac loading warning
- Add missing "altr,sysmgr-syscon" property to all gmac nodes

* tag 'stratix10_fix_for_v5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/dinguyen/linux:
  arm64: dts: stratix10: add the sysmgr-syscon property from the gmac's

Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoMerge tag 'reset-fixes-for-v5.1' of git://git.pengutronix.de/pza/linux into arm/fixes
Olof Johansson [Sun, 7 Apr 2019 22:14:00 +0000 (15:14 -0700)]
Merge tag 'reset-fixes-for-v5.1' of git://git.pengutronix.de/pza/linux into arm/fixes

Reset controller fixes for v5.1

This tag adds missing USB PHY reset lines to the Meson G12A reset
controller header and fixes the Meson Audio ARB driver to prevent
module unloading while it is in use.

* tag 'reset-fixes-for-v5.1' of git://git.pengutronix.de/pza/linux:
  reset: meson-audio-arb: Fix missing .owner setting of reset_controller_dev
  dt-bindings: reset: meson-g12a: Add missing USB2 PHY resets

Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agodt-bindings: cpu: Fix JSON schema
Maxime Ripard [Mon, 18 Mar 2019 10:05:21 +0000 (11:05 +0100)]
dt-bindings: cpu: Fix JSON schema

Commit fd73403a4862 ("dt-bindings: arm: Add SMP enable-method for
Milbeaut") added support for a new cpu enable-method, but did so using
tabulations to ident. This is however invalid in the syntax, and resulted
in a failure when trying to use that schemas for validation.

Use spaces instead of tabs to indent to fix this.

Fixes: fd73403a4862 ("dt-bindings: arm: Add SMP enable-method for Milbeaut")
Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Acked-by: Sugaya Taichi <sugaya.taichi@socionext.com>
Signed-off-by: Olof Johansson <olof@lixom.net>
5 years agoMerge tag 'for-linus-5.1b-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 7 Apr 2019 16:12:10 +0000 (06:12 -1000)]
Merge tag 'for-linus-5.1b-rc4-tag' of git://git./linux/kernel/git/xen/tip

Pull xen fixes from Juergen Gross:
 "One minor fix and a small cleanup for the xen privcmd driver"

* tag 'for-linus-5.1b-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen: Prevent buffer overflow in privcmd ioctl
  xen: use struct_size() helper in kzalloc()

5 years agoMerge tag 'mtd/fixes-for-5.1-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 7 Apr 2019 16:07:20 +0000 (06:07 -1000)]
Merge tag 'mtd/fixes-for-5.1-rc4' of git://git./linux/kernel/git/mtd/linux

Pull MTD fix from Richard Weinberger:
 "A single fix for a possible infinite loop in the cfi_cmdset_0002
  driver"

* tag 'mtd/fixes-for-5.1-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux:
  mtd: cfi: fix deadloop in cfi_cmdset_0002.c do_write_buffer

5 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sun, 7 Apr 2019 16:00:35 +0000 (06:00 -1000)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Five small fixes. Four in three drivers: qedi, lpfc and storvsc. The
  final one is labelled core, but merely adds a dh rdac entry for Lenovo
  systems"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: lpfc: Fix missing wakeups on abort threads
  scsi: storvsc: Reduce default ring buffer size to 128 Kbytes
  scsi: storvsc: Fix calculation of sub-channel count
  scsi: core: add new RDAC LENOVO/DE_Series device
  scsi: qedi: remove declaration of nvm_image from stack

5 years agonfc: nci: Potential off by one in ->pipes[] array
Dan Carpenter [Wed, 3 Apr 2019 07:13:51 +0000 (10:13 +0300)]
nfc: nci: Potential off by one in ->pipes[] array

This is similar to commit e285d5bfb7e9 ("NFC: Fix the number of pipes")
where we changed NFC_HCI_MAX_PIPES from 127 to 128.

As the comment next to the define explains, the pipe identifier is 7
bits long.  The highest possible pipe is 127, but the number of possible
pipes is 128.  As the code is now, then there is potential for an
out of bounds array access:

    net/nfc/nci/hci.c:297 nci_hci_cmd_received() warn: array off by one?
    'ndev->hci_dev->pipes[pipe]' '0-127 == 127'

Fixes: 11f54f228643 ("NFC: nci: Add HCI over NCI protocol support")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoNFC: nci: Add some bounds checking in nci_hci_cmd_received()
Dan Carpenter [Wed, 3 Apr 2019 07:12:48 +0000 (10:12 +0300)]
NFC: nci: Add some bounds checking in nci_hci_cmd_received()

This is similar to commit 674d9de02aa7 ("NFC: Fix possible memory
corruption when handling SHDLC I-Frame commands").

I'm not totally sure, but I think that commit description may have
overstated the danger.  I was under the impression that this data came
from the firmware?  If you can't trust your networking firmware, then
you're already in trouble.

Anyway, these days we add bounds checking where ever we can and we call
it kernel hardening.  Better safe than sorry.

Fixes: 11f54f228643 ("NFC: nci: Add HCI over NCI protocol support")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'i2c/for-current-fixed' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 6 Apr 2019 21:52:59 +0000 (11:52 -1000)]
Merge branch 'i2c/for-current-fixed' of git://git./linux/kernel/git/wsa/linux

Pull i2c fix from Wolfram Sang:
 "A simple but wanted driver bugfix"

* 'i2c/for-current-fixed' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: imx: don't leak the i2c adapter on error

5 years agoMerge branch 'parisc-5.1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Sat, 6 Apr 2019 20:59:30 +0000 (10:59 -1000)]
Merge branch 'parisc-5.1-2' of git://git./linux/kernel/git/deller/parisc-linux

Pull parisc fixes from Helge Deller:
 "A 32-bit boot regression fix introduced in the merge window, a QEMU
  detection fix and two fixes by Sven regarding ptrace & kprobes"

* 'parisc-5.1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Detect QEMU earlier in boot process
  parisc: also set iaoq_b in instruction_pointer_set()
  parisc: regs_return_value() should return gpr28
  Revert: parisc: Use F_EXTEND() macro in iosapic code

5 years agoparisc: Detect QEMU earlier in boot process
Helge Deller [Tue, 2 Apr 2019 10:13:27 +0000 (12:13 +0200)]
parisc: Detect QEMU earlier in boot process

While adding LASI support to QEMU, I noticed that the QEMU detection in
the kernel happens much too late. For example, when a LASI chip is found
by the kernel, it registers the LASI LED driver as well.  But when we
run on QEMU it makes sense to avoid spending unnecessary CPU cycles, so
we need to access the running_on_QEMU flag earlier than before.

This patch now makes the QEMU detection the fist task of the Linux
kernel by moving it to where the kernel enters the C-coding.

Fixes: 310d82784fb4 ("parisc: qemu idle sleep support")
Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org # v4.14+
5 years agoparisc: also set iaoq_b in instruction_pointer_set()
Sven Schnelle [Thu, 4 Apr 2019 16:16:04 +0000 (18:16 +0200)]
parisc: also set iaoq_b in instruction_pointer_set()

When setting the instruction pointer on PA-RISC we also need
to set the back of the instruction queue to the new offset, otherwise
we will execute on instruction from the new location, and jumping
back to the old location stored in iaoq_b.

Signed-off-by: Sven Schnelle <svens@stackframe.org>
Signed-off-by: Helge Deller <deller@gmx.de>
Fixes: 75ebedf1d263 ("parisc: Add HAVE_REGS_AND_STACK_ACCESS_API feature")
Cc: stable@vger.kernel.org # 4.19+
5 years agoparisc: regs_return_value() should return gpr28
Sven Schnelle [Thu, 4 Apr 2019 16:16:03 +0000 (18:16 +0200)]
parisc: regs_return_value() should return gpr28

While working on kretprobes for PA-RISC I was wondering while the
kprobes sanity test always fails on kretprobes. This is caused by
returning gpr20 instead of gpr28.

Signed-off-by: Sven Schnelle <svens@stackframe.org>
Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org # 4.14+
5 years agoRevert: parisc: Use F_EXTEND() macro in iosapic code
Helge Deller [Mon, 18 Mar 2019 21:56:15 +0000 (22:56 +0100)]
Revert: parisc: Use F_EXTEND() macro in iosapic code

Revert parts of commit 97d7e2e3fd8a ("parisc: Use F_EXTEND() macro in
iosapic code"). It breaks booting the 32-bit kernel on some machines.

Reported-by: Sven Schnelle <svens@stackframe.org>
Tested-by: Sven Schnelle <svens@stackframe.org>
Fixes: 97d7e2e3fd8a ("parisc: Use F_EXTEND() macro in iosapic code")
Signed-off-by: Helge Deller <deller@gmx.de>
5 years agofs: stream_open - opener for stream-like files so that read and write can run simulta...
Kirill Smelkov [Tue, 26 Mar 2019 22:20:43 +0000 (22:20 +0000)]
fs: stream_open - opener for stream-like files so that read and write can run simultaneously without deadlock

Commit 9c225f2655e3 ("vfs: atomic f_pos accesses as per POSIX") added
locking for file.f_pos access and in particular made concurrent read and
write not possible - now both those functions take f_pos lock for the
whole run, and so if e.g. a read is blocked waiting for data, write will
deadlock waiting for that read to complete.

This caused regression for stream-like files where previously read and
write could run simultaneously, but after that patch could not do so
anymore. See e.g. commit 581d21a2d02a ("xenbus: fix deadlock on writes
to /proc/xen/xenbus") which fixes such regression for particular case of
/proc/xen/xenbus.

The patch that added f_pos lock in 2014 did so to guarantee POSIX thread
safety for read/write/lseek and added the locking to file descriptors of
all regular files. In 2014 that thread-safety problem was not new as it
was already discussed earlier in 2006.

However even though 2006'th version of Linus's patch was adding f_pos
locking "only for files that are marked seekable with FMODE_LSEEK (thus
avoiding the stream-like objects like pipes and sockets)", the 2014
version - the one that actually made it into the tree as 9c225f2655e3 -
is doing so irregardless of whether a file is seekable or not.

See

    https://lore.kernel.org/lkml/53022DB1.4070805@gmail.com/
    https://lwn.net/Articles/180387
    https://lwn.net/Articles/180396

for historic context.

The reason that it did so is, probably, that there are many files that
are marked non-seekable, but e.g. their read implementation actually
depends on knowing current position to correctly handle the read. Some
examples:

kernel/power/user.c snapshot_read
fs/debugfs/file.c u32_array_read
fs/fuse/control.c fuse_conn_waiting_read + ...
drivers/hwmon/asus_atk0110.c atk_debugfs_ggrp_read
arch/s390/hypfs/inode.c hypfs_read_iter
...

Despite that, many nonseekable_open users implement read and write with
pure stream semantics - they don't depend on passed ppos at all. And for
those cases where read could wait for something inside, it creates a
situation similar to xenbus - the write could be never made to go until
read is done, and read is waiting for some, potentially external, event,
for potentially unbounded time -> deadlock.

Besides xenbus, there are 14 such places in the kernel that I've found
with semantic patch (see below):

drivers/xen/evtchn.c:667:8-24: ERROR: evtchn_fops: .read() can deadlock .write()
drivers/isdn/capi/capi.c:963:8-24: ERROR: capi_fops: .read() can deadlock .write()
drivers/input/evdev.c:527:1-17: ERROR: evdev_fops: .read() can deadlock .write()
drivers/char/pcmcia/cm4000_cs.c:1685:7-23: ERROR: cm4000_fops: .read() can deadlock .write()
net/rfkill/core.c:1146:8-24: ERROR: rfkill_fops: .read() can deadlock .write()
drivers/s390/char/fs3270.c:488:1-17: ERROR: fs3270_fops: .read() can deadlock .write()
drivers/usb/misc/ldusb.c:310:1-17: ERROR: ld_usb_fops: .read() can deadlock .write()
drivers/hid/uhid.c:635:1-17: ERROR: uhid_fops: .read() can deadlock .write()
net/batman-adv/icmp_socket.c:80:1-17: ERROR: batadv_fops: .read() can deadlock .write()
drivers/media/rc/lirc_dev.c:198:1-17: ERROR: lirc_fops: .read() can deadlock .write()
drivers/leds/uleds.c:77:1-17: ERROR: uleds_fops: .read() can deadlock .write()
drivers/input/misc/uinput.c:400:1-17: ERROR: uinput_fops: .read() can deadlock .write()
drivers/infiniband/core/user_mad.c:985:7-23: ERROR: umad_fops: .read() can deadlock .write()
drivers/gnss/core.c:45:1-17: ERROR: gnss_fops: .read() can deadlock .write()

In addition to the cases above another regression caused by f_pos
locking is that now FUSE filesystems that implement open with
FOPEN_NONSEEKABLE flag, can no longer implement bidirectional
stream-like files - for the same reason as above e.g. read can deadlock
write locking on file.f_pos in the kernel.

FUSE's FOPEN_NONSEEKABLE was added in 2008 in a7c1b990f715 ("fuse:
implement nonseekable open") to support OSSPD. OSSPD implements /dev/dsp
in userspace with FOPEN_NONSEEKABLE flag, with corresponding read and
write routines not depending on current position at all, and with both
read and write being potentially blocking operations:

See

    https://github.com/libfuse/osspd
    https://lwn.net/Articles/308445

    https://github.com/libfuse/osspd/blob/14a9cff0/osspd.c#L1406
    https://github.com/libfuse/osspd/blob/14a9cff0/osspd.c#L1438-L1477
    https://github.com/libfuse/osspd/blob/14a9cff0/osspd.c#L1479-L1510

Corresponding libfuse example/test also describes FOPEN_NONSEEKABLE as
"somewhat pipe-like files ..." with read handler not using offset.
However that test implements only read without write and cannot exercise
the deadlock scenario:

    https://github.com/libfuse/libfuse/blob/fuse-3.4.2-3-ga1bff7d/example/poll.c#L124-L131
    https://github.com/libfuse/libfuse/blob/fuse-3.4.2-3-ga1bff7d/example/poll.c#L146-L163
    https://github.com/libfuse/libfuse/blob/fuse-3.4.2-3-ga1bff7d/example/poll.c#L209-L216

I've actually hit the read vs write deadlock for real while implementing
my FUSE filesystem where there is /head/watch file, for which open
creates separate bidirectional socket-like stream in between filesystem
and its user with both read and write being later performed
simultaneously. And there it is semantically not easy to split the
stream into two separate read-only and write-only channels:

    https://lab.nexedi.com/kirr/wendelin.core/blob/f13aa600/wcfs/wcfs.go#L88-169

Let's fix this regression. The plan is:

1. We can't change nonseekable_open to include &~FMODE_ATOMIC_POS -
   doing so would break many in-kernel nonseekable_open users which
   actually use ppos in read/write handlers.

2. Add stream_open() to kernel to open stream-like non-seekable file
   descriptors. Read and write on such file descriptors would never use
   nor change ppos. And with that property on stream-like files read and
   write will be running without taking f_pos lock - i.e. read and write
   could be running simultaneously.

3. With semantic patch search and convert to stream_open all in-kernel
   nonseekable_open users for which read and write actually do not
   depend on ppos and where there is no other methods in file_operations
   which assume @offset access.

4. Add FOPEN_STREAM to fs/fuse/ and open in-kernel file-descriptors via
   steam_open if that bit is present in filesystem open reply.

   It was tempting to change fs/fuse/ open handler to use stream_open
   instead of nonseekable_open on just FOPEN_NONSEEKABLE flags, but
   grepping through Debian codesearch shows users of FOPEN_NONSEEKABLE,
   and in particular GVFS which actually uses offset in its read and
   write handlers

https://codesearch.debian.net/search?q=-%3Enonseekable+%3D
https://gitlab.gnome.org/GNOME/gvfs/blob/1.40.0-6-gcbc54396/client/gvfsfusedaemon.c#L1080
https://gitlab.gnome.org/GNOME/gvfs/blob/1.40.0-6-gcbc54396/client/gvfsfusedaemon.c#L1247-1346
https://gitlab.gnome.org/GNOME/gvfs/blob/1.40.0-6-gcbc54396/client/gvfsfusedaemon.c#L1399-1481

   so if we would do such a change it will break a real user.

5. Add stream_open and FOPEN_STREAM handling to stable kernels starting
   from v3.14+ (the kernel where 9c225f2655 first appeared).

   This will allow to patch OSSPD and other FUSE filesystems that
   provide stream-like files to return FOPEN_STREAM | FOPEN_NONSEEKABLE
   in their open handler and this way avoid the deadlock on all kernel
   versions. This should work because fs/fuse/ ignores unknown open
   flags returned from a filesystem and so passing FOPEN_STREAM to a
   kernel that is not aware of this flag cannot hurt. In turn the kernel
   that is not aware of FOPEN_STREAM will be < v3.14 where just
   FOPEN_NONSEEKABLE is sufficient to implement streams without read vs
   write deadlock.

This patch adds stream_open, converts /proc/xen/xenbus to it and adds
semantic patch to automatically locate in-kernel places that are either
required to be converted due to read vs write deadlock, or that are just
safe to be converted because read and write do not use ppos and there
are no other funky methods in file_operations.

Regarding semantic patch I've verified each generated change manually -
that it is correct to convert - and each other nonseekable_open instance
left - that it is either not correct to convert there, or that it is not
converted due to current stream_open.cocci limitations.

The script also does not convert files that should be valid to convert,
but that currently have .llseek = noop_llseek or generic_file_llseek for
unknown reason despite file being opened with nonseekable_open (e.g.
drivers/input/mousedev.c)

Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Yongzhi Pan <panyongzhi@gmail.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: Tejun Heo <tj@kernel.org>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Julia Lawall <Julia.Lawall@lip6.fr>
Cc: Nikolaus Rath <Nikolaus@rath.org>
Cc: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Kirill Smelkov <kirr@nexedi.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoxsysace: Fix error handling in ace_setup
Guenter Roeck [Tue, 19 Feb 2019 16:49:56 +0000 (08:49 -0800)]
xsysace: Fix error handling in ace_setup

If xace hardware reports a bad version number, the error handling code
in ace_setup() calls put_disk(), followed by queue cleanup. However, since
the disk data structure has the queue pointer set, put_disk() also
cleans and releases the queue. This results in blk_cleanup_queue()
accessing an already released data structure, which in turn may result
in a crash such as the following.

[   10.681671] BUG: Kernel NULL pointer dereference at 0x00000040
[   10.681826] Faulting instruction address: 0xc0431480
[   10.682072] Oops: Kernel access of bad area, sig: 11 [#1]
[   10.682251] BE PAGE_SIZE=4K PREEMPT Xilinx Virtex440
[   10.682387] Modules linked in:
[   10.682528] CPU: 0 PID: 1 Comm: swapper Tainted: G        W         5.0.0-rc6-next-20190218+ #2
[   10.682733] NIP:  c0431480 LR: c043147c CTR: c0422ad8
[   10.682863] REGS: cf82fbe0 TRAP: 0300   Tainted: G        W          (5.0.0-rc6-next-20190218+)
[   10.683065] MSR:  00029000 <CE,EE,ME>  CR: 22000222  XER: 00000000
[   10.683236] DEAR: 00000040 ESR: 00000000
[   10.683236] GPR00: c043147c cf82fc90 cf82ccc0 00000000 00000000 00000000 00000002 00000000
[   10.683236] GPR08: 00000000 00000000 c04310bc 00000000 22000222 00000000 c0002c54 00000000
[   10.683236] GPR16: 00000000 00000001 c09aa39c c09021b0 c09021dc 00000007 c0a68c08 00000000
[   10.683236] GPR24: 00000001 ced6d400 ced6dcf0 c0815d9c 00000000 00000000 00000000 cedf0800
[   10.684331] NIP [c0431480] blk_mq_run_hw_queue+0x28/0x114
[   10.684473] LR [c043147c] blk_mq_run_hw_queue+0x24/0x114
[   10.684602] Call Trace:
[   10.684671] [cf82fc90] [c043147c] blk_mq_run_hw_queue+0x24/0x114 (unreliable)
[   10.684854] [cf82fcc0] [c04315bc] blk_mq_run_hw_queues+0x50/0x7c
[   10.685002] [cf82fce0] [c0422b24] blk_set_queue_dying+0x30/0x68
[   10.685154] [cf82fcf0] [c0423ec0] blk_cleanup_queue+0x34/0x14c
[   10.685306] [cf82fd10] [c054d73c] ace_probe+0x3dc/0x508
[   10.685445] [cf82fd50] [c052d740] platform_drv_probe+0x4c/0xb8
[   10.685592] [cf82fd70] [c052abb0] really_probe+0x20c/0x32c
[   10.685728] [cf82fda0] [c052ae58] driver_probe_device+0x68/0x464
[   10.685877] [cf82fdc0] [c052b500] device_driver_attach+0xb4/0xe4
[   10.686024] [cf82fde0] [c052b5dc] __driver_attach+0xac/0xfc
[   10.686161] [cf82fe00] [c0528428] bus_for_each_dev+0x80/0xc0
[   10.686314] [cf82fe30] [c0529b3c] bus_add_driver+0x144/0x234
[   10.686457] [cf82fe50] [c052c46c] driver_register+0x88/0x15c
[   10.686610] [cf82fe60] [c09de288] ace_init+0x4c/0xac
[   10.686742] [cf82fe80] [c0002730] do_one_initcall+0xac/0x330
[   10.686888] [cf82fee0] [c09aafd0] kernel_init_freeable+0x34c/0x478
[   10.687043] [cf82ff30] [c0002c6c] kernel_init+0x18/0x114
[   10.687188] [cf82ff40] [c000f2f0] ret_from_kernel_thread+0x14/0x1c
[   10.687349] Instruction dump:
[   10.687435] 3863ffd4 4bfffd70 9421ffd0 7c0802a6 93c10028 7c9e2378 93e1002c 38810008
[   10.687637] 7c7f1b78 90010034 4bfffc25 813f008c <8129004075290100 4182002c 80810008
[   10.688056] ---[ end trace 13c9ff51d41b9d40 ]---

Fix the problem by setting the disk queue pointer to NULL before calling
put_disk(). A more comprehensive fix might be to rearrange the code
to check the hardware version before initializing data structures,
but I don't know if this would have undesirable side effects, and
it would increase the complexity of backporting the fix to older kernels.

Fixes: 74489a91dd43a ("Add support for Xilinx SystemACE CompactFlash interface")
Acked-by: Michal Simek <michal.simek@xilinx.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agonull_blk: prevent crash from bad home_node value
John Pittman [Fri, 5 Apr 2019 21:42:45 +0000 (17:42 -0400)]
null_blk: prevent crash from bad home_node value

At module load, if the selected home_node value is greater than
the available numa nodes, the system will crash in
__alloc_pages_nodemask() due to a bad paging request.  Prevent this
user error crash by detecting the bad value, logging an error, and
setting g_home_node back to the default of NUMA_NO_NODE.

Signed-off-by: John Pittman <jpittman@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoMerge tag 'rtc-5.1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux
Linus Torvalds [Sat, 6 Apr 2019 16:26:36 +0000 (06:26 -1000)]
Merge tag 'rtc-5.1-2' of git://git./linux/kernel/git/abelloni/linux

Pull RTC fixes from Alexandre Belloni:

 - Various alarm fixes for da9063, cros-ec and sh

 - sd3078 manufacturer name fix as this was introduced this cycle

* tag 'rtc-5.1-2' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux:
  rtc: da9063: set uie_unsupported when relevant
  rtc: sd3078: fix manufacturer name
  rtc: sh: Fix invalid alarm warning for non-enabled alarm
  rtc: cros-ec: Fail suspend/resume if wake IRQ can't be configured

5 years agoi2c: imx: don't leak the i2c adapter on error
Laurentiu Tudor [Mon, 1 Apr 2019 10:14:37 +0000 (13:14 +0300)]
i2c: imx: don't leak the i2c adapter on error

Make sure to free the i2c adapter on the error exit path.

Signed-off-by: Laurentiu Tudor <laurentiu.tudor@nxp.com>
Reviewed-by: Mukesh Ojha <mojha@codeaurora.org>
Reviewed-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Fixes: e1ab9a468e3b ("i2c: imx: improve the error handling in i2c_imx_dma_request()")
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
5 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sat, 6 Apr 2019 03:08:55 +0000 (17:08 -1000)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "14 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  kernel/sysctl.c: fix out-of-bounds access when setting file-max
  mm/util.c: fix strndup_user() comment
  sh: fix multiple function definition build errors
  MAINTAINERS: add maintainer and replacing reviewer ARM/NUVOTON NPCM
  MAINTAINERS: fix bad pattern in ARM/NUVOTON NPCM
  mm: writeback: use exact memcg dirty counts
  psi: clarify the units used in pressure files
  mm/huge_memory.c: fix modifying of page protection by insert_pfn_pmd()
  hugetlbfs: fix memory leak for resv_map
  mm: fix vm_fault_t cast in VM_FAULT_GET_HINDEX()
  lib/lzo: fix bugs for very short or empty input
  include/linux/bitrev.h: fix constant bitrev
  kmemleak: powerpc: skip scanning holes in the .bss section
  lib/string.c: implement a basic bcmp

5 years agokernel/sysctl.c: fix out-of-bounds access when setting file-max
Will Deacon [Sat, 6 Apr 2019 01:39:38 +0000 (18:39 -0700)]
kernel/sysctl.c: fix out-of-bounds access when setting file-max

Commit 32a5ad9c2285 ("sysctl: handle overflow for file-max") hooked up
min/max values for the file-max sysctl parameter via the .extra1 and
.extra2 fields in the corresponding struct ctl_table entry.

Unfortunately, the minimum value points at the global 'zero' variable,
which is an int.  This results in a KASAN splat when accessed as a long
by proc_doulongvec_minmax on 64-bit architectures:

  | BUG: KASAN: global-out-of-bounds in __do_proc_doulongvec_minmax+0x5d8/0x6a0
  | Read of size 8 at addr ffff2000133d1c20 by task systemd/1
  |
  | CPU: 0 PID: 1 Comm: systemd Not tainted 5.1.0-rc3-00012-g40b114779944 #2
  | Hardware name: linux,dummy-virt (DT)
  | Call trace:
  |  dump_backtrace+0x0/0x228
  |  show_stack+0x14/0x20
  |  dump_stack+0xe8/0x124
  |  print_address_description+0x60/0x258
  |  kasan_report+0x140/0x1a0
  |  __asan_report_load8_noabort+0x18/0x20
  |  __do_proc_doulongvec_minmax+0x5d8/0x6a0
  |  proc_doulongvec_minmax+0x4c/0x78
  |  proc_sys_call_handler.isra.19+0x144/0x1d8
  |  proc_sys_write+0x34/0x58
  |  __vfs_write+0x54/0xe8
  |  vfs_write+0x124/0x3c0
  |  ksys_write+0xbc/0x168
  |  __arm64_sys_write+0x68/0x98
  |  el0_svc_common+0x100/0x258
  |  el0_svc_handler+0x48/0xc0
  |  el0_svc+0x8/0xc
  |
  | The buggy address belongs to the variable:
  |  zero+0x0/0x40
  |
  | Memory state around the buggy address:
  |  ffff2000133d1b00: 00 00 00 00 00 00 00 00 fa fa fa fa 04 fa fa fa
  |  ffff2000133d1b80: fa fa fa fa 04 fa fa fa fa fa fa fa 04 fa fa fa
  | >ffff2000133d1c00: fa fa fa fa 04 fa fa fa fa fa fa fa 00 00 00 00
  |                                ^
  |  ffff2000133d1c80: fa fa fa fa 00 fa fa fa fa fa fa fa 00 00 00 00
  |  ffff2000133d1d00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00

Fix the splat by introducing a unsigned long 'zero_ul' and using that
instead.

Link: http://lkml.kernel.org/r/20190403153409.17307-1-will.deacon@arm.com
Fixes: 32a5ad9c2285 ("sysctl: handle overflow for file-max")
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Christian Brauner <christian@brauner.io>
Cc: Kees Cook <keescook@chromium.org>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matteo Croce <mcroce@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>